Categories

Is your bank casting the wrong spell with its AI magic wand? (Andrew Beatty)

How well do you really know your bank’s customers? Have you met them at all? Do you know where and how they shop, what their interests and aspirations are, or their general state of mind and personal circumstances?

You should. But of course, no one has psychoanalysis tools or magical algorithms to find such data on each consumer, to see exactly where each cent of their money goes to help them in their financial lives. But this data is important. Why? Consumers are desperate for financial help. The problem has been building for a long time, but with the COVID-19 pandemic, financial health concerns moved to the forefront. Costs keep rising while paychecks stay stagnant (or were lost) and many businesses are working hard to stay afloat.

Neither you nor I have magical powers, but artificial intelligence (AI) can help us come close… For banks, AI could serve as your “magic wand” to know and address consumer needs. But there is a caveat: AI will only work if you collect the right data with the right intent.

There is a lot of data, but where is wisdom?

Over the last decade fintech emerged out of banks’ inability to offer a unified experience and unbundled individual services. Every banking function is being torn down to its most fundamental version. Transaction monitoring, peer-to-peer (P2P) payments, Buy Now Pay Later (BNPL) – you name it. Investors are pouring money into unbundling services, and customers have embraced them big time. From a consumer lens, it is fantastic, but is it good for AI insights?

With the simplicity of unbundling comes the complexity of data aggregation. When each financial services function is deconstructed, the data is too. Data now sits across a multitude of platforms, and that makes it even more challenging for banks to construct AI models without creating blind spots that result in biased or inaccurate predictions. So, as more data sources continue to form, we must ensure that we don’t sacrifice fairness and accuracy in the name of quick results. We must ensure that we address our data blind spots, so that inequality is not exacerbated.

Banks need to reflect on their true core.

Undoubtedly, financial services should be unbundled and chopped down to fundamentals, but the data needs to be consolidated. This is easier said than done, and with GDPR and CCPA regulations data sharing from multiple sources becomes even more challenging and expensive. But if banks look deep down, their real strength sits right at their core. In literal terms your core banking system can support an AI design project. Though the core does not contain all the data about customers, it has enough data to start building an equitable AI system. Data scientists may argue that small data will create imbalance, difficulty in optimization etc. but in such a scenario where data is wide, small data is a much better source to start.

What should banks do?

In simple terms you are looking at a 2-step process: Banks should start building models based on the available transaction data and – most importantly – train the models to also identify the missing data. The next step – which is the heart of the solution – is to communicate with your consumers. One moral that we’ve learned from the pandemic is that banks have lost touch with their customers … and AI cannot function without communication.

Based on the limited predictions and identified blind spots in the data, your bank can contextually communicate with consumers using modern-day bots or conversational AI to validate and calibrate the missing data. Performing validity checks with the data source and the owner of the data for bias at various points in the model development process becomes more crucial to design AI models that anticipate interactions and correct biases. These can be the magical steps that will enable your bank to design an AI system that is fast and inexpensive, and which can comprehensively understand and assist your customers. This is true human-centered AI.

Many financial institutions have invested millions in search of an AI magic wand, but have not reaped true rewards yet. That is not due to the “AI wand” – it’s because they’re casting the wrong spell on the right wand. That can change, now!

Source